Finish #9484: a single-card leaves-the-battlefield trigger checks look-back for itself

[haksanlulz]

Finish #9484: a single-card leaves-the-battlefield trigger checks look-back for itself

Fixes #9484

Summary

A card exiled "until this leaves the battlefield" comes back during the host's own zone change (Card.runLeavesPlayCommands() inside GameAction.changeZone), so its triggers are registered before the host's dies trigger finds its watchers. In #9484 a Slaughter Specialist exiled by Brutal Cathar returned when Brutal Cathar died, saw it die, and got a +1/+1 counter.

It shouldn't. Brutal Cathar's ruling (2021-09-24): "The exiled card returns to the battlefield immediately after this creature leaves the battlefield. Nothing happens between the two events, including state-based actions. The two creatures aren't on the battlefield at the same time." A dies trigger looks back in time (CR 603.10a), so it sees the game as it was just before the event, with the Specialist still in exile.

#9491 ("Part for #9484") handles this wherever the death goes through a zone table's batch event: CardZoneTable.triggerChangesZoneAll rebuilds the active triggers from the event's last-state battlefield, so a permanent that is in play now but absent from that snapshot is left out. This covers the other part, a ChangesZone trigger evaluated without that batch event, where the trigger itself had no look-back check while TriggerChangesZoneAll.performTest already has one.

What changed

• TriggerChangesZone.performTest makes the check TriggerChangesZoneAll already makes: a trigger that works on the battlefield, for an event whose origin is the battlefield, doesn't fire when its host is in play now but isn't in the event's last-state battlefield. A delayed trigger (no active zone) is left alone, and a host that has itself left is not in play, so its own leaves trigger, registered from its last known information, still fires.
• The check reads AbilityKey.LastStateBattlefield from the run params, which effects (AbilityKey.addCardZoneTableParams), costs and state-based actions already pass, and does nothing when the params carry none.

Verification

• UntilHostLeavesLookBackTest (forge-gui-desktop, 3 tests): Brutal Cathar exiles a watcher, a second copy of the watcher enters as the control, Brutal Cathar dies. The copy on the battlefield gets 1 counter, the returned one gets 0. (1) The death arrives through destroy with a zone table and no batch event: on master it fails with expected:<0> but was:<1>, here it passes. (2) Murder cast from hand, then state-based actions, for both Slaughter Specialist (ChangesZone) and Sengir Connoisseur (ChangesZoneAll): guards Only register LTB trigger that are in lastStateBattlefield #9491's prune. (3) The other direction: Zombie Apocalypse returns Midnight Reaper and then destroys a Human in the same resolution, and the Reaper still takes its 1 damage, so the check does not suppress a permanent that entered after the resolution's snapshot was taken.
• mvn -U -B clean -P windows-linux install, Java 17, on this branch (based on master bb00615): BUILD SUCCESS, 0 Checkstyle violations, 773 tests, 0 failures, 6 skipped (NetworkPlayIntegrationTest).

Size

136+ / 0- across 2 files, one commit; 126 of the additions are the test.

Disclosure

Written with an AI coding agent (Claude Code); tested as described above. The commit carries the co-author line.

🤖 Generated with Claude Code

[Hanmac]

should probably be combined with the block below at line 111-117?

[haksanlulz]

I folded them in, thanks.

[tool4ever]

I don't know, this isn't how it'd happen in the engine normally and that's the only reason you get this test to fail...

[haksanlulz]

That’s fair. I think it's still worth keeping the route tests, #9491 has none and they’d be the only thing that notices if the prune ever regresses. The check itself I can keep as is now that it’s folded, or drop, either works. The original issue can be closed in anycase.